Alibaba launched Qwen 3.7 in July 2026 — a 72-billion-parameter reasoning model priced at $0.6/M input and $2/M output. It continues Alibaba’s strategy: near-frontier performance at budget pricing with open weights.
What’s New
Qwen 3.7 improves reasoning efficiency over its predecessor — stronger on math and logic benchmarks per token of compute — while keeping the open-weight license that made the Qwen family a self-hosting favorite.
Pricing Advantage
At $0.6/$2, Qwen 3.7 undercuts GPT-5.6 Luna ($1/$6) and sits near DeepSeek V3.2 ($0.5/$2). For high-volume workloads, the savings are substantial.
Open-Weight Landscape
Qwen joins Kimi K3 (2.8T) and Llama 4 (405B) in the open-weights race. The Asian labs are winning the price-performance battle; the question is whether quality catches up to the closed frontier.
Who Should Use It
Budget-constrained teams, self-hosters with data sovereignty needs, and high-volume structured workloads. Benchmark against DeepSeek and Llama on your exact tasks — see the AI Model Comparison 2026 for the comparison framework.
